Heart Sound Signal Quality Assessment Based on Multi-Domain Features

Heart sound is one of the most important physiological signals of our body, including a large number of physiological and pathological information that can reflect the cardiovascular status. This study aims to develop a heart sound signal quality assessment method. In view of the 3 common noises (deep breath, speaking and cough) in clinical data collection, a total of 72 features were extracted from 6 domains, i.e., time, frequency, entropy, energy, high-order statistics and cyclostationarity. Then information gain, which was used as feature selection method, as well as statistical analysis were employed for dimension reduction. A SVM with radial basis kernel function was trained for final signal quality classification. The best effect was obtained on distinguishing resting from cough and the result showed that the classification performance was significantly improved after feature selection. In contrast, statistical analysis had little effect on the improvement of classification results. The best accuracy in distinguishing between resting and deep breath, resting and speaking, resting and cough is 87.73%, 95.00%, 98.64%, respectively. These results indicate that the proposed method is effective for identifying different noise states, namely cough, speaking and deep breath.